What I'm listening to - March 2025

Here's a roundup of what grabbed my attention this past month. This month featured debates around AI’s capabilities and blind spots. Plus a great interview with an maniacal podcast founder.

Apple Earnings, OpenAI Deep Research, The Unbundling of Substantiation - Stratechery
- Ben Thompson creates two Deep Research reports about Apple Earnings. One prompt is generic (i.e. "I want a research report about Apple's latest earnings") and the other include specific angles that he wants to explore.
- It's fascinating that the second example is so much better. AI excels at fleshing out well-defined ideas but struggles with generating original insights. Ben calls this out as the "unbundling" of the creation and substantiation of an idea.
An Interview with Benedict Evans About AI Unknowns - Stratechery
- This Stratechery interview pairs nicely with the above and teases out some more subtle ideas.

Evans also tried Deep Research and found that the example on the OpenAI website to analyze smartphone adoption was incorrect in a few ways. First, it failed to define clearly what "adoption" meant. Second, it pulled from a weak source. And third, it used a different number than what was in the source.

It's a great example of how LLMs give people superpowers but do have limits. Right now, they perform best with qualitative work that doesn't have a precise answer (e.g. "What’s good to bring to a picnic?"). In other words, "LLMs are good at the things that computers are bad at, and bad at the things that computers are good at."

I Met Charlie Munger and Discovered How Billionaires Really Think - The Startup Ideas Podcast
- I liked learning about how David Senra thinks more than the Charlie Munger stories. He's obsessed with reading biographies and just as obsessed with distilling the important lessons. "I mark things up. I highlight them. I add them to a giant database."
- Dan Carlin's hardcore way of recording Hardcore History "I thought it was scripted and it's like, no, he's going to read 30 books, right? He's going to spend half a year doing that. And he's going to sit down and he records it little by little. So he'll record, you know, little segments over half a year, nine months, and then stitches them all together." Old tweet here.
- Senra says "Biographies are the closest thing to finding a cheat code in real life". Great example from Elon Musk: "I didn't have a mentor... I read biographies"

Some Charlie Munger wisdom that deeply resonates: "I didn't succeed because of intelligence. I succeeded because I had a long attention span"

Chris Pedregal - Building Granola - Invest Like The Best
- I shared a longer list of AI tools I've been digging, but Granola is one of the AI products that I use every day. It's a meeting transcriber that does a phenomenal job of generating notes and identifying action items. Granola eliminates my note taking anxiety. I can just focus on the conversation and let the app do its thing.

This was a great interview to learn a little bit more about the Granola founder and how they approach shipping great product. Related to the Deep Research discussion, he's clearly thought a lot about when LLMs are useful "I think what's incredible about LLMs is that you can use them to bring extremely relevant context to the person in the moment they need it."

The interview asks why ChatGPT or another frontier lab won't recreate Granola. Chris makes a good point that power tools are all about the overall experience and the bespoke UX is just as big of an unlock as AI: "It's not a question of intelligence. It's actually how great is the UI optimized for this use case. And I think that if you have a product that is solely dedicated to being phenomenal at that use case, it will be a better experience than a general tool will be."
I wish they pushed further on this point. LLMs are already really good at building UIs and iterating on them. Chris does point out that things are moving really quickly and what we will outsource to AI will be completely different in 3 years.